If you upgrade to 2.0.2 (I don't understand why that would be difficult)
It is difficult at this point because we are already doing testing. As a temp solution, we set the chunk size to 1. It solved the unwanted transaction rollback and reprocessing issues. Still, I'm posting these questions because I'm not happy with that approach. It is not really 'by batch' per se.

Probably next time I'll try the 2.0.2 with no-rollback-for and propagation=NOT_SUPPORTED. My question now is, is this the recommended approach or standard practice for a use case like mine?

IMO, my use case is quite simple and straightforward but the Batch solution is not that simple and straightforward.